Zero-shot downstream evaluation
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
74.82ARC-Easy Accuracy
26
May 21, 2026
56.57ARC-E Accuracy
14
May 8, 2026
Downstream Tasks MMLU, HellaSwag, PIQA, BoolQ, WinoGrande, ARC-E, ARC-C, OBQA (500 samples per task)
74.4MMLU
7
Apr 28, 2026
50.1HellaSwag Accuracy
4
May 20, 2026